WO2021248554A1

WO2021248554A1 - High-speed and high-precision burr detection method and detection system for lithium ion battery electrode sheet

Info

Publication number: WO2021248554A1
Application number: PCT/CN2020/097452
Authority: WO
Inventors: 周华民; 刘家欢; 黄天仓; 杨志明; 张云; 黄志高
Original assignee: 深圳市信宇人科技股份有限公司
Priority date: 2020-06-11
Filing date: 2020-06-22
Publication date: 2021-12-16
Also published as: CN111650210A; CN111650210B

Abstract

A high-speed and high-precision burr detection method and detection system for a lithium ion battery electrode sheet (15). The method comprises: S1, electrifying and initializing a detection system; S2, at the beginning of detection, a high-speed line scanning camera (12) carrying out high-frequency sampling on electrode sheets (15); S31, the detection system reading, from a memory and at predetermined time intervals, images collected by the line scanning camera (12); S32, while the detection system reads scanning images and carries out detection, the line scanning camera (12) collecting images of the electrode sheets (15) in parallel and storing the collected images in the memory, and waiting for the next instance of reading; S4, calculating the length of a defect post-processing buffer area of an electrode sheet (15) according to a winding linear speed v of the electrode sheet (15) and the total consumed time ta of algorithm detection, and after a burr defect is detected by means of algorithm detection, suspending winding or marking processing according to requirements; and S5, repeating steps S31-S4 until all the electrode sheets (15) are tested. The detection method and detection system can meet production requirements when slitting is carried out at high speed and a blur is tiny in size.

Description

Burr detection method and detection system for high-speed and high-precision lithium ion battery pole pieces

Technical field

The invention belongs to the technical field of visual inspection, and specifically relates to a burr detection method and a matching detection system for a burr in a lithium ion battery pole piece slitting process.

Background technique

Lithium-ion batteries are currently the most widely used power source in the world due to their large capacity and high energy efficiency ratio. As the performance requirements of lithium-ion batteries are getting higher and higher, the accuracy requirements for each process of battery manufacturing have also become higher. The production process of the pole piece of the lithium ion battery is a key link, and the quality control is very strict. During the slitting process of lithium battery pole pieces, defects such as burrs will be generated on the current collector of the pole pieces. The burrs can pierce the battery diaphragm during battery assembly, cause the battery to be short-circuited and discarded, and even cause safety problems. Therefore, the detection of burr defects of lithium-ion battery pole pieces is very important.

The existing burr detection method simply uses an industrial camera to shoot the image of the pole piece, and then the burr detection is performed by the detection algorithm. However, due to the limitation of the frame rate and sampling accuracy of industrial cameras, it cannot meet the requirements of high-speed and high-precision burr detection. In the case of high-speed slitting and small burr size, the existing detection technology cannot meet the production requirements. This requires a well-arranged light source structure to light the pole piece, and then design a suitable image acquisition system and image processing system to complete the high-speed and high-precision pole piece burr detection requirements. Summary of the invention

technical problem

The solution to the problem

Technical solutions

In view of the above described problems, the present invention provides a high-speed and high-precision burr detection method and detection system for lithium-ion battery pole pieces based on a modular image acquisition system composed of a herringbone structure light source and a high-speed line scan camera; It has the characteristics that it can meet the requirements of production under the condition of high-speed slitting and small burr size.

The herringbone structure light source of the present invention is a customized light source matched with the camera used in the present invention. Two common surface light sources similar to strip light sources contact at one end to form a herringbone structure. The light source and the camera form a modular image acquisition Module. In addition, the angle of this herringbone structure can be adjusted to suit different shooting scenes. Further, in order to improve the quality of image collection, a flat backlight is placed opposite to the chevron light source at the lower part of the edge portions on both sides of the pole piece of the lithium battery to improve the contrast of the pole piece. The image acquisition device in the present invention is a high-speed line scan camera, and the installation direction is perpendicular to the movement direction of the pole piece. The high-speed scanning of the line scan camera is used to match the high-speed movement of the pole piece during winding. At the same time, because the line frame resolution of general line scan cameras can reach up to 16384 pixels, it can meet the requirements of high-precision burr detection.

The deep learning-based defect detection algorithm proposed by the present invention uses Convolutional Neural Network (CNN) to process input images, extract image features, and classify and locate defect regions.

In order to achieve the above objectives, the present invention also proposes a high-speed and high-precision lithium-ion battery pole piece burr visual inspection system, which includes mechanical structures such as cameras, light sources, industrial computers, and corresponding light source controllers, as well as matching detection algorithms. Among them:

The camera is a high-speed line scan camera, and the line frame resolution needs to be selected according to the specific use scene. For example, the width of the battery pole piece involved in the present invention is about 10-20cm after the slitting process, and the size of the burr is about 5μm. In order to ensure the detection accuracy, the defect area should be described by 3-5 pixels, so the pixels of the camera The resolution should reach 1pixel/μm. In order to save cost and improve efficiency, this system uses two above-mentioned cameras and light source modules to detect burr defects on both sides of the pole piece. The FOV of the camera used in this system is 10mm. Therefore, a line scan camera with a line frame resolution of 12K is selected in this system. The speed of the pole piece during rewinding is about 1m/s, so the sampling frequency of the camera should reach ¹⁰⁶ lines/sec. This system uses a scanning camera with a vertical resolution of 4 pixels and a sampling frequency of 300KHz. Therefore, the line scan camera used in this system has a resolution of 12K×4 pixels and a frequency of 300KHz to meet the needs of high-speed and high-precision detection.

The said camera is equipped with an ordinary optical lens, and a lens matched with the working distance of the camera is adopted in this system;

The said herringbone light source is a pure-color strip-shaped surface light source. White light is used in this system. The light source controller used in conjunction with the light source can perform 256-level brightness dimming with the light source.

The plane backlight source is a pure-color surface light source, and white light is used in this system. The light source controller used in conjunction with the light source can perform 256-level brightness dimming with the light source.

The burr detection system also includes a mechanical structure part, specifically a fixed fixture of the image acquisition device, a light source angle adjustment mechanical structure, and a pole piece air-jet cleaning pretreatment mechanism;

The described burr detection algorithm uses a CNN-based defect detection framework, preprocesses the image output by the line scan camera and inputs it into the detection framework, performs target detection, and outputs the defect type and location. Specifically, according to different use environments and requirements, the model in the algorithm can also be trained to detect defects such as scratches, inclusions, and leaking foils.

The described CNN-based defect detection framework requires offline training before actual use. The specific method is as follows:

(1) Implement the described CNN-based defect detection model framework;

(2) Collect a predetermined amount (preferably 200-300) of pole piece images containing defects, and a professional will mark the position of the defect with a rectangular frame, and give the defect category to obtain a training data sample;

(3) Input the training data samples into the CNN model for feature extraction calculation, then classify the defects in the figure and regress to calculate the position of the defect, and give the probability that the suspected area is the type of defect;

(4) Compare the calculation result of the model in step (3) with the result of the actual manual marking, calculate the calculation error of the model, and use the back propagation rule to update the parameters of the model, thereby realizing the optimization of the model.

After the offline training of the CNN model framework is completed, the model parameters are called by the matching detection system, and the defect area is predicted online during actual detection.

The described defect detection algorithm process is to obtain the image output by the line scan camera, first perform the image average filter processing on the image to reduce the effect of noise in the image on the result, and then use the gray linear change to improve the defect area and background area in the image Contrast to highlight defects for easy detection. Then, the processed image is scaled to a fixed size and then passed into the above-mentioned CNN model for defect classification and location prediction.

Preferably, the detection method of the lithium battery pole piece burr detection system includes the following steps:

S1. The light source and image acquisition system are installed on the pole piece winding assembly line, and the light source and camera are adjusted well to obtain a good pole piece sampling image, power on, and initialize the detection system;

S2. At the beginning of the detection, the high-speed line scan camera performs high-frequency sampling on the pole pieces cleaned by the air jet, and calculates the corresponding sampling frequency according to the winding line speed, so that the camera sampling frequency is synchronized with the pole piece movement speed;

S31. The detection system reads the line scan camera image from the memory at a predetermined time interval, and then preprocesses the complete image after stitching the scan camera line frame, uses the mean filter to smooth the noise of the input image, and then executes the linear gray scale transformation to highlight Defect area, and then input the preprocessed result image into the CNN defect detection model for burr detection, and complete the detection in a very short time;

S32. While the system reads the scanned image and detects it, the camera still collects the pole piece in parallel at this time and stores it in the memory, waiting for the next reading;

S4. It is easy to calculate the length of the post-processing buffer for pole piece defects according to the pole piece winding line speed v and the total time t _{a of} _{the detection algorithm, the length is L c} =v*t _a , the defect detection algorithm is After the burr defect is detected, the rewinding or marking process shall be suspended as required. The algorithm calculates the relative position of the burr and processes the burr in the buffer;

S5. Repeat steps S31 to S4 until all pole pieces are tested.

Preferably, the CNN defect detection model is modified according to the YOLO v3 framework. Because the content information in the industrial defect image is less, and the requirement of high-speed inspection is taken into consideration, the original YOLO v3 framework is streamlined and modified to adapt to the defect inspection task. Specifically, the CNN defect detection model includes a backbone feature extraction network VGG16 model and three additional convolution branches for predicting results. The result of the backbone network VGG16 contains 5 consecutive building blocks consisting of a convolutional layer, a pooling layer and an activation function. These 5 building blocks have a total of 13 layers, which are used to extract features from the image. In order to improve the accuracy of the model's detection of small defect areas and reduce the calculation time, the additional 3 predictive convolution branches are directly derived from the main VGG16 network to detect defects on multi-scale features. Specifically, the first predictive convolution branch is derived from the fifth building block of the VGG16 backbone network, and features are extracted through a 3×3 convolution layer, and then a 1×1 convolution layer predicts the result. Among them, the features output by the fifth layer after being used to predict the results also pass through the first upsampling layer, are expanded by a factor of 2 and then merged with the features from the fourth building block, and then enter the second prediction convolution branch, after a The 3×3 convolutional layer extracts features, and then a 1×1 convolutional layer predicts the result. Similarly, the merged feature passes through the second up-sampling layer, and is expanded by a factor of 2 and then merged with the feature from the third building block. Then enter the third prediction convolution branch, extract features through a 3×3 convolution layer, and then predict the result by a 1×1 convolution layer. Predicting the results after three different scale features improves the model's detection accuracy of defects, especially for small defect types, such as leaking foils.

Preferably, since the types of defects of the lithium battery pole pieces are not many and the changes are small, each predicted convolution branch predicts 3 different suspected defect region candidate frames at each pixel position of the feature map.

In order to ensure the imaging quality and detection accuracy, preliminary debugging of the entire system is required. The first is the hardware part, including the adjustment of the camera's installation position and working distance to obtain a clear and complete pole piece edge image; at the same time, the brightness and angle of the herringbone light source are adjusted to cooperate with the camera to adjust to obtain a good sample image; The second part is the system, including the image reading interval, camera parameter settings, and exposure parameters to coordinate with the high-speed winding pole piece movement to obtain high-quality sampled images.

The beneficial effects of the invention

Beneficial effect

In general, compared with the prior art through the above technical solutions conceived by the present invention, the lithium battery pole piece burr detection method and system based on the high-speed line scan camera and the herringbone structure light source provided by the present invention mainly have the following beneficial effects :

(1) The light source with herringbone structure can well light the surface of the lithium battery pole piece according to the sampling characteristics of the camera to obtain a clear image. It forms a light source pair with the flat backlight under the pole piece to make burr defects The pole piece background is easy to distinguish;

(2) The adopted high-speed line scan camera can cope with the scene of precise detection of tiny burrs in the process of high-speed slitting and winding of pole pieces. The speed of s is as small as 5μm for burr defect detection, and the burr detection accuracy and recall rate can reach more than 99%, so as to realize the high-speed and high-precision detection of lithium battery pole pieces.

(3) The described image acquisition module has the advantage of modularity. The hardware structure and the matching detection software system can flexibly adapt to the detection of pole piece burrs of different widths, which improves the efficiency and has better applicability and economy. .

Brief description of the drawings

Description of the drawings

Fig. 1 is a schematic diagram of an image acquisition module composed of a high-speed line scan camera and a herringbone light source according to the present invention; Fig. 2 is a schematic top view of the structure of Fig. 1;

FIG. 3 is a schematic diagram of the optical path structure of the image acquisition module described in FIG. 1;

4 is a schematic diagram of the top view structure of the hardware layout of the detection system according to the present invention;

Figure 5 is a schematic diagram of the A-direction structure in Figure 4;

FIG. 6 is a schematic diagram of the block structure of the detection system according to the present invention;

FIG. 7 is a schematic diagram of the block structure of the detection algorithm according to the present invention;

FIG. 8 is a schematic diagram of the structure of the defect detection frame according to the present invention;

Fig. 9 is a schematic diagram of a screenshot of the detection result of the detection system used in the present invention.

The best embodiment for implementing the invention

The best mode of the present invention

In order to make the objectives, technical solutions, and advantages of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, but not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

Please refer to Figures 1 to 5. Figures 1 to 5 disclose a high-speed and high-precision burr detection system for lithium-ion battery pole pieces, including an image acquisition module 1 and a detection module 2. The image acquisition module 1 includes a light source 11 And a line scan camera 12, the light source 11 is used to illuminate a predetermined part of the pole piece, the line scan camera 12 scans and photographs the illuminated part of the pole piece, and transmits the captured image to the detection module 2.

Preferably, the detection module 2 processes the captured images according to the following high-speed and high-precision burr detection method of lithium-ion battery pole pieces.

Preferably, the light source 11 includes a first light source 111 and a second light source 112, the first light source 111 is arranged on the first mounting member 113; the second light source 112 is arranged on the second mounting member 114, the The upper ends of the first mounting member 113 and the second mounting member 114 are pivotally connected by a shaft 115 to form a herringbone light source. The line scan camera 12 is arranged on the center line of the herringbone light source, and the line scan camera 12 The lens is aimed at the irradiation area of the light source 11.

Preferably, on the side far from the light source 11, a flat backlight 13 is further provided under the pole piece 15, and the light emitting direction of the flat backlight 13 faces the bottom surface of the pole piece; it is used to distinguish between burr defects and Pole piece background.

Preferably, there are two image acquisition modules 1, which are respectively located on both sides of the pole piece in the length direction, and are used to collect burrs on both sides of the pole piece.

Preferably, the present invention further includes an air-jet cleaning device 14 which is located on the upstream side of the image acquisition module 1 and is used to remove dust from the pole pieces.

As shown in Figure 6, the burr detection system scheme of the present invention is that after the hardware installation and debugging of the system is completed, the supporting detection software completes the initialization of all components, and the industrial computer controls the herringbone structure light source to reach the predetermined brightness through the light source controller. The machine is connected to multiple high-speed line scan cameras through the switch. After the image acquisition module is sampled, the software system calls the algorithm to complete the glitch detection, and feeds back the detection result to the execution end to complete the predetermined action.

With reference to Figure 7, the detection steps of the present invention are explained as follows:

S1. Debug the position of the light source, the angle of the light source, the installation position of the camera, the working distance, and set the parameters of the image acquisition module. At the same time, install the air-jet cleaning device and the processing execution end of the buffer zone to the predetermined position;

S2. When the detection starts, the detection system software uses the PLC to control the air-jet cleaning device to remove impurities on the surface of the lithium battery pole pieces to prevent the impurities from affecting the subsequent burr detection;

S31. The image acquisition module performs pole piece image sampling at a set frequency, acquires the pole piece surface image, and stores it in the memory for later use. Two image acquisition modules distinguish and collect edge images on both sides of the pole piece;

S32. The detection software system obtains the images obtained in step S31 at regular intervals, first stitches the images obtained by line scanning into a large image, and then merges the images on both sides obtained by the two image modules into one image, and then matches The defect detection software system first performs filtering and noise reduction processing on the image, and then uses the image grayscale transformation algorithm to adjust the contrast of the pole image to highlight the defect area. After the preprocessed image is obtained, it is sent to the CNN detection framework to detect the burr in the image. At the same time, the image module is still synchronized to continue to collect images for backup;

S4. Since the detection algorithm takes a certain amount of time, the area after the last image acquisition and detection has entered the set buffer processing area, and the system detection result is fed back to the execution end through the PLC, and the scheduled shutdown check or other mark is performed Work;

S5. Repeat steps S31 to S4 until all pole pieces are tested.

Preferably, as shown in FIG. 8, the convolution kernel parameters of each layer of the CNN structure model used in the present invention are shown in the following table. Among them, Conv stands for convolutional layer.

The CNN model uses the VGG16 structure as the backbone network, so in the offline training stage, the migration learning method is used to use the pre-trained model on the public image data set in the defect image training process of the present invention to improve the accuracy of model detection. , Reduce the training time and the number of training samples required.

It is easy for those skilled in the art to understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement and improvement, etc. made within the spirit and principle of the present invention, All should be included in the protection scope of the present invention.

Claims

A high-speed and high-precision burr detection method for lithium-ion battery pole pieces includes the following steps:

S1, power on, initialize the detection system;

S2. At the beginning of the detection, the high-speed line scan camera pole piece performs high-frequency sampling, and the corresponding sampling frequency is calculated according to the winding line speed, so that the camera sampling frequency is synchronized with the pole piece movement speed;

S31. The detection system reads the line scan camera image from the memory at a predetermined time interval, and then preprocesses the complete image after stitching the scan camera line frame, uses the mean filter to smooth the noise of the input image, and then executes the linear gray scale transformation to highlight Defect area, and then input the preprocessed result image into the CNN defect detection model for burr detection, and complete the detection in a very short time;

S32. While the inspection system reads and inspects the scanned image, the line scan camera simultaneously collects and stores the image of the pole piece in the memory, and waits for the next reading;

S4. Calculate the length of the post-processing buffer for the pole piece defect according to the pole piece winding line speed v and the total time ta of the detection algorithm, the length is Lc=v*ta, the defect detection algorithm detects the burr defect , Pause rewinding or mark processing as required; the algorithm calculates the relative position of the burr, and processes the burr in the buffer;

S5. Repeat steps S31 to S4 until all pole pieces are tested.
The burr detection method for high-speed and high-precision lithium-ion battery pole pieces according to claim 1, wherein the CNN defect detection model includes a backbone feature extraction network VGG16 model and three additional volumes for predicting results The composition of the product branch; the backbone network VGG16 result contains 5 continuous building blocks composed of convolutional layer, pooling layer and activation function; these 5 building blocks have 13 layers, which are used to extract features from the image; in order to improve the model For the detection accuracy of small defect areas, while reducing the calculation time, the additional 3 prediction convolution branches are directly derived from the main VGG16 network to detect defects on multi-scale features.
The burr detection method for high-speed and high-precision lithium-ion battery pole pieces according to claim 2, wherein the first predictive convolution branch is derived from the fifth building module of the main feature extraction network VGG16 model, and passes through a 3 ×3 convolutional layer extracts the features, and then a 1×1 convolutional layer predicts the result; among them, the features output by the 5th layer are used in the prediction result and go through the first up-sampling layer. The features of the two building modules are fused, and then enter the second prediction convolution branch, through a 3×3 convolutional layer to extract the features, and then a 1×1 convolutional layer to predict the result; the fused feature then passes through the second The up-sampling layer is expanded by 2 times and then merged with the features from the third building block; then it enters the third prediction convolution branch, passes through a 3×3 convolution layer to extract features, and then a 1×1 convolution layer forecast result.
The burr detection method for high-speed and high-precision lithium-ion battery pole pieces according to claim 3, wherein each predictive convolution branch predicts three different suspected defective area candidate frames at each pixel position of the feature map .
A high-speed and high-precision burr detection system for lithium-ion battery pole pieces, which is characterized in that it includes an image acquisition module (1) and a detection module (2). The image acquisition module (1) includes a light source (11) and a line scan camera (12) The light source (11) is used to illuminate a predetermined part of the pole piece, and the line scan camera (12) scans and photographs the illuminated part of the pole piece, and transmits the captured image to the Detection module (2).
The burr detection system for high-speed and high-precision lithium-ion battery pole pieces according to claim 5, wherein the detection module (2) is based on the high-speed and high-precision lithium-ion battery pole of any one of claims 1-4. The burr detection method of the film processes the captured images.
The burr detection system for high-speed and high-precision lithium-ion battery pole pieces according to claim 5 or 6, wherein the light source (11) comprises a first light source (111) and a second light source (112), and the first light source (111) and the second light source (112) A light source (111) is provided on the first mounting member (113); the second light source (112) is provided on the second mounting member (114), and the first mounting member (113) and the second mounting member ( The upper end of 114) is pivotally connected by a shaft (115) to form a herringbone light source, the line scan camera (12) is set on the center line of the herringbone light source, and the lens of the line scan camera (12) is aligned The irradiation area of the light source (11).
The burr detection system for high-speed and high-precision lithium-ion battery pole pieces according to claim 5 or 6, characterized in that, on the side far from the light source (11), a flat backlight (13) is provided under the pole pieces. ), the light emitting direction of the flat backlight (13) faces the bottom surface of the pole piece; it is used to distinguish burr defects from the pole piece background.
The high-speed and high-precision burr detection system for lithium-ion battery pole pieces according to claim 5 or 6, characterized in that there are two image acquisition modules (1), which are respectively located on both sides of the length of the pole piece for Collect the burrs on both sides of the pole piece.
The burr detection system for high-speed and high-precision lithium-ion battery pole pieces according to claim 5 or 6, characterized in that it further comprises an air-jet cleaning device (14), and the air-jet cleaning device (14) is located in the image acquisition module ( The upstream side of 1) is used to remove dust from the pole pieces.